We Got Slopped in 2025 Ai Inside #107 - CourtneyR.dev's 2nd Brain

We Got Slopped in 2025 // Ai Inside #107 🔗 **Source:** [We Got Slopped in 2025 // Ai Inside #107](https://www.youtube.com/live/pJ6P_ECUZkI) > [!note]+ Document Notes > tags: [AI-podcast, tech-industry, AI-consolidation, anti-AI-movement, AI-acquisitions] > author: AI Inside > source: YouTube/Podcast > reference: https://www.youtube.com/live/pJ6P_ECUZkI > date created: 2025-01-02 10:30 > > # 🏷️ Title: We Got Slopped in 2025 // AI Inside #107 >  ## 🧰 Action Inbox - [H1] Monitor anti-AI sentiment impact on political campaigns — emerging voter sentiment #politics - [H2] Evaluate AI authenticity vs efficiency in business processes — customer trust concerns #business - [H3] Track consolidation trends in AI inference market — strategic positioning #strategy ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [New York Times: An Anti-AI Movement is Coming](https://www.nytimes.com) - [Politico: Americans Hate AI](https://www.politico.com) ## 📝 Summary - Major AI industry consolidation with Nvidia licensing Groq technology and Meta acquiring Manis for $2 billion - Growing anti-AI sentiment creating unusual political coalitions across traditional party lines - Personal experiences reveal AI authenticity challenges in customer-facing applications - Industry moving toward inference-focused strategies as training models commoditize - Year-end reflection on AI's rapid development and societal impact ## 📖 Main Content This episode covers two major AI acquisitions and the emerging backlash against AI technology. Nvidia made a strategic move to license Groq's inference chip technology while hiring key executives, positioning itself to dominate both training and inference markets. Meta acquired AI agent startup Manis for $2 billion, bringing proven revenue-generating AI agents into its ecosystem. The hosts discuss growing anti-AI sentiment, noting how it's creating unusual political coalitions with Trump and some Democrats supporting AI companies while Bernie Sanders and some Republicans push for regulation. Personal anecdotes reveal the authenticity challenges AI creates in customer interactions. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/pJ6P_ECUZkI?t=0) — [H1] Episode introduction and sponsor acknowledgment - Key idea: Year-end episode focusing on major AI consolidation and backlash - Evidence: "We are squeezing in a quick episode before the end of the year" #### ⏱️ [02:30](https://youtu.be/pJ6P_ECUZkI?t=150) — [H2] Nvidia-Groq deal strategic implications - Key idea: Nvidia is strategically positioning for inference market dominance - Evidence: "Jensen Wong is betting that the models are trained, the models are commodified, the models are going to change" #### ⏱️ [05:45](https://youtu.be/pJ6P_ECUZkI?t=345) — [H3] Groq's SRAM-based technology advantage - Key idea: Groq's chips offer extremely fast access through integrated memory - Evidence: "Fairly small amount of memory but as part of the processor so it's extremely fast access" #### ⏱️ [08:20](https://youtu.be/pJ6P_ECUZkI?t=500) — [H2] Unusual deal structure preserves independence - Key idea: License-plus-talent acquisition differs from traditional buyouts - Evidence: "First reports were that he bought the company... but instead was a license company stays out there" #### ⏱️ [12:15](https://youtu.be/pJ6P_ECUZkI?t=735) — [H1] Meta acquires Manis for $2 billion - Key idea: Meta purchases proven AI agent company with $100M ARR - Evidence: "Fastest startup to go from $0 to $100 million in the world because only at eight months since launch" #### ⏱️ [15:30](https://youtu.be/pJ6P_ECUZkI?t=930) — [H3] Manis background and capabilities - Key idea: Singapore-based startup excelled at practical AI agent tasks - Evidence: "Showing off their agents doing things like screening job candidates, planning vacations, analyzing portfolios" #### ⏱️ [18:45](https://youtu.be/pJ6P_ECUZkI?t=1125) — [H2] Political implications of AI backlash - Key idea: Anti-AI sentiment creates unusual cross-party coalitions - Evidence: "Trump and some Democratic leaders supporting AI companies... Republican leaders and Bernie Sanders together pushing for more regulation" #### ⏱️ [22:10](https://youtu.be/pJ6P_ECUZkI?t=1330) — [H3] Historical context for technology backlash - Key idea: Current AI resistance follows predictable patterns from past technologies - Evidence: "Section 230 was bipartisan... then others from both parties smashing section 230" #### ⏱️ [25:20](https://youtu.be/pJ6P_ECUZkI?t=1520) — [H2] Personal experience with AI authenticity challenges - Key idea: AI-generated customer communications feel inauthentic to recipients - Evidence: "The email felt very inauthentic... doesn't sound like you" #### ⏱️ [28:40](https://youtu.be/pJ6P_ECUZkI?t=1720) — [H3] Getting "slopped" by AI content - Key idea: Even AI-aware users fall for sophisticated AI-generated content - Evidence: "I got slopped... it was too perfect. It was too this. It was too that" #### ⏱️ [32:15](https://youtu.be/pJ6P_ECUZkI?t=1935) — [H2] AI as collaborative writing tool - Key idea: AI works better as educational partner than content creator - Evidence: "More useful as an educational exercise for me... I understood what was wrong with it" #### ⏱️ [35:50](https://youtu.be/pJ6P_ECUZkI?t=2150) — [H3] OpenAI's preparedness role offering - Key idea: High-compensation safety role reflects company's risk acknowledgment - Evidence: "This will be a stressful job and you'll jump into the deep end pretty much immediately" ## 📚 References - [Nate Jones YouTube Channel](https://youtube.com/@natejonesai) - [Your360.ai](https://your360.ai) - [Groq AI Technology](https://groq.com) - [Manis AI Agents](https://manis.ai) ## 🗣️ Paraphrase - Nvidia strategically licensed Groq's inference technology and hired key executives to dominate both AI training and inference markets - Meta acquired Manis for $2 billion, bringing proven AI agent technology with significant revenue into their platform ecosystem - Growing anti-AI sentiment is creating unusual political coalitions that cross traditional party lines in unexpected ways ## 📰 Newsletter **Major AI Consolidation Shapes 2025 Landscape** The AI industry closed 2024 with significant consolidation moves that will impact the entire ecosystem. Nvidia's strategic licensing of Groq's inference technology positions them to dominate both training and inference markets, while Meta's $2 billion Manis acquisition brings proven AI agents to billions of users. Meanwhile, growing anti-AI sentiment is creating interesting political dynamics, with traditional party lines blurring around AI regulation. Personal experiences with AI authenticity challenges remind us that technical capability doesn't always equal user acceptance. **Key Takeaways:** - Strategic licensing deals may become preferred alternative to full acquisitions - AI inference market becoming as important as training chip market - User authenticity concerns growing alongside AI capabilities Resources: [AI Inside Podcast](https://aiinsideshow.com), [Groq Technology](https://groq.com), [Your360.ai](https://your360.ai) ## ✅ Action Items #### DevRel - Developer advocacy: Monitor community sentiment around AI authenticity and trust - Education: Create guidance on balancing AI efficiency with authentic communication - Developer experience: Document best practices for AI-human hybrid workflows - Community management: Address growing concerns about AI-generated content #### OSPO - Policy & Governance: Track regulatory proposals from both parties on AI development - License Compliance: Review implications of strategic licensing deals vs acquisitions - Security Management: Assess risks of consolidated AI infrastructure - Strategic Alignment: Monitor inference market evolution for organizational planning #### Community - Maintainers: Consider community reaction to AI-assisted development tools - Contributors: Balance AI productivity gains with authentic human interaction - Sponsors: Evaluate messaging around AI use in sponsored content #### WordPress - Core: Assess implications of AI consolidation for open source CMS ecosystem - Community: Monitor sentiment around AI tools in WordPress development - Training: Develop guidance on transparent AI use in educational content - Marketing: Balance efficiency gains with authentic community communication ## 🧪 Hypotheses - Strategic licensing will become preferred alternative to full acquisitions in AI space - Anti-AI sentiment will significantly influence 2026 midterm election messaging - Inference market consolidation will reshape AI cost structures and accessibility - User authenticity concerns will drive demand for AI disclosure standards ### 📊 Methodology - Survey analysis of voter priorities including AI regulation preferences - Market analysis of inference vs training chip demand trends - User experience studies on AI authenticity perception - Political coalition mapping around AI policy positions ### ❓ Questions - Will inference market consolidation limit AI accessibility for smaller companies? - How will cross-party AI coalitions affect traditional political alignments? - Can AI systems achieve authenticity or will human oversight remain essential? ### 🔮 Future Research - Long-term political implications of AI backlash movements - Economic impact of inference market consolidation - User acceptance patterns for AI-generated vs human content ### 💡 Implications - Practice: Balance AI efficiency with authentic human communication in customer interactions - Policy: Prepare for bipartisan AI regulation that may slow development - Security: Monitor concentration risks in AI infrastructure - Education: Develop literacy around detecting and disclosing AI-generated content - Funding: Strategic licensing deals may require new valuation frameworks ### 🧠 Learning Styles Assessment Reading level estimate: 12th Grade (Flesch-Kincaid Grade Level) #### 👀 Visual learners - Present: YouTube video format with host interaction - Improve: Add market consolidation diagrams, technology architecture visuals #### 🔊 Aural learners - Present: Full audio podcast format with conversational discussion - Improve: Chapter markers for easier navigation to key topics #### 🗣️ Verbal learners - Present: Clear definitions of technical terms like "inference" and "SRAM" - Improve: More analogies explaining complex AI infrastructure concepts #### 🧩 Kinesthetic learners - Present: Real-world examples and personal anecdotes - Improve: Interactive demos of technologies discussed #### 🧠 Logical learners - Present: Strategic analysis of business implications - Improve: Timeline graphics showing consolidation patterns #### 🧑‍🤝‍🧑 Social learners - Present: Dialog format between co-hosts with different perspectives - Improve: Community discussion prompts in show notes #### 🧘‍♀️ Solitary learners - Present: Individual reflection opportunities through personal anecdotes - Improve: Self-assessment tools for AI use in business contexts #### 🌿 Naturalistic learners - Present: Technology adoption patterns and evolution analogies - Improve: Ecosystem thinking about AI industry interdependencies ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Audio format naturally accessible to screen readers - Improve: Detailed show notes with proper heading structure #### Keyboard navigation - Present: Standard YouTube player controls - Improve: Timestamp links for easy navigation to specific topics #### Contrast and readability - Present: Audio-first format reduces visual accessibility barriers - Improve: High contrast show notes and transcript formatting #### Media and captions - Present: Full transcript available from YouTube auto-generation - Improve: Professional transcription with speaker identification ### 🌍 DEIB Assessment #### Cultural breadth - Present: Global perspective on AI companies from China, Singapore, US - Improve: More diverse geographic examples and implications #### Inclusive language - Present: Technical jargon explained in accessible terms - Improve: Reduce assumptions about technical background #### Neurodiversity support - Present: Conversational format with natural processing time - Improve: Content warnings for rapid technical discussions #### Representation - Present: Discussion of diverse business models and approaches - Improve: More varied perspectives beyond tech industry insiders ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear logical progression from market consolidation to strategic implications - Inductive strength: Pattern recognition from historical technology adoption cycles - Causal reasoning: Links between user authenticity concerns and AI backlash - Analogical strength: Comparison to past technology regulation cycles #### Fallacies - Appeal to authority: Some reliance on "Jensen Wong is strategic" without detailed analysis - Anecdotal evidence: Personal experiences used to generalize broader trends #### Tone of voice - Overall tone: Conversational and analytical, balancing enthusiasm with realistic assessment ### 🛡️ DARVO Analysis - Deny: No denial patterns observed - Attack: No attack patterns observed - Reverse victim/offender: No reversal patterns observed - Emotional cues and rhetoric: Balanced discussion without emotional manipulation ## 🎯 Conclusion The AI industry's consolidation through strategic licensing deals and major acquisitions signals a maturing market focused on practical applications and inference capabilities. However, growing public skepticism and political backlash may significantly impact future development and regulation. The tension between AI efficiency and human authenticity represents a critical challenge for organizations implementing these technologies. Success will likely depend on transparent integration that preserves human connection while leveraging AI capabilities. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad ### [H1] Strategic licensing emerges as acquisition alternative > [!quote] We Got Slopped in 2025 // AI Inside #107 by [[AI Inside]] via [[YouTube]]. Published 2024-12-30. Accessed 2025-01-02. > "Rather than saying well we got to build this ourselves he said no they're good they're out there and he went after them" Nvidia's approach to Groq demonstrates new model for AI consolidation. [view highlight](https://youtu.be/pJ6P_ECUZkI?t=500#H1) #xref #acquisition ID:rw_nvidia_groq ### [H2] Anti-AI sentiment crosses traditional political lines > [!quote] We Got Slopped in 2025 // AI Inside #107 > "Trump and some Democratic leaders supporting AI companies while Bernie Sanders and Republicans push for regulation" Political coalitions around AI don't follow expected patterns. [view highlight](https://youtu.be/pJ6P_ECUZkI?t=1125#H2) #xref #politics ID:rw_ai_politics --- ### 🔖 Atomic Split Pad (Zettel Candidates) #### [Z1] Inference markets become strategic AI battleground - Scope: Post-training AI chip market evolution - Pointer: 02:30 timestamp discussion - Links: [[AI-Infrastructure]] [[Market-Consolidation]] #### [Z2] AI authenticity challenges user acceptance - Scope: Customer-facing AI implementation concerns - Pointer: 25:20 personal anecdote - Links: [[User-Experience]] [[AI-Trust]] --- ### ✏️ Fleeting Queue - [F1] Research Nate Jones's chain of verification methodology — YouTube channel #fleeting #research - [F2] Track Bernie Sanders AI moratorium proposal — political implications #fleeting #politics - [F3] Monitor Groq independence post-Nvidia --- ## Full Transcript [00:01] This episode of the AI Inside podcast is [00:04] sponsored by your 360.ai. [00:07] Get 10% off through January 2026 by [00:10] using code inside. [00:21] [music] [00:24] Hello everybody. Happy New Year almost [00:26] like in a couple of days. I'm one of [00:29] your hosts of AI Inside, Jason How. This [00:31] is the show where we take a look at the [00:32] AI that is layered throughout the world [00:34] of technology. Joined as always by Jeff [00:36] Jarvis. Good to see you, Jeff. [00:38] >> Hello, boss. How are you doing? All [00:40] right. We are squeezing in a quick [00:42] episode um before the end of the year. [00:46] Tomorrow be, you know, at the time of [00:47] this recording, tomorrow is New Year's [00:49] Eve. I was like, do we want to do a [00:51] podcast on New Year's Eve? Turns out I'm [00:53] doing another thing on New Year's Eve [00:55] anyways, so it's not like I get the day [00:56] off. [00:56] >> Party boy, huh? [laughter] [00:58] No, not even. No, it's not that. But it [01:02] is, you know, it's sometimes you just [01:04] got to take the opportunities when they [01:05] come. [01:05] >> And you think think maybe people are [01:07] going to watch, but last week's episode [01:08] with Ben Girtzil um is is is steaming on [01:12] YouTube. So that was the Christmas Eve [01:14] episode as it turned out. [01:15] >> Got to say doing an awesome uh amount of [01:19] traffic and which I kind of had a [01:21] feeling was going to be the case. Ben is [01:22] a pretty in demand kind of figure when [01:26] it comes to the world of AI. So if you [01:28] have not checked last week's episode, we [01:30] obviously we did not have a news episode [01:32] in the feed, but we did have our [01:34] interview with Ben Girtzil and uh you [01:37] know it went really well. Ben's a [01:39] fascinating character, fascinating [01:41] individual in the world of technology [01:43] and in the world of AI. Literally coined [01:45] the term uh artificial general [01:47] intelligence 20 some odd years ago. So, [01:49] he's an important voice and uh that is [01:52] evidenced by the fact that we've had a [01:54] lot of people tuning into the interview [01:56] at least on YouTube. So, uh thanks again [01:58] to the Singularity Net team for allowing [02:01] us to chat with Ben. [02:03] >> Yep. [02:03] >> Yeah. Making it happen. And uh I'm sure [02:05] we're going to have lots more wonderful [02:07] interviews coming up in the new year. Um [02:10] so, we've got some news today to talk [02:12] about. [02:13] >> I'm going to had no news, quiet week [02:15] between nothing happening, but we got [02:16] some big stories actually. Hey, I mean [02:18] it just keeps on going and yes, so we've [02:20] got a couple of big stories more in the [02:22] like acquisition aqua hire kind of side [02:25] of things and a few other things and [02:27] then near the end of the episode instead [02:29] of doing our normal like speed round [02:31] thing that we've been doing figured we'd [02:33] take a little time to just kind of do a [02:35] little reflection. You know, it's the [02:36] last episode of the year. Why not look [02:38] at 2025 and AI [snorts] [02:41] >> and not not at like a cohesive like this [02:44] happened and then this and then this [02:45] happened. We're too lazy for that. [02:47] >> Yeah. Yeah. It's [02:48] >> it's too it's too much to look back on. [02:50] Geez. [cough] [02:51] >> It's [clears throat] It's so much. [02:52] That's the problem. So, I thought why [02:55] don't we just kind of pick out a few [02:57] things that really stood out to us. So, [02:59] that's what we're going to close on. But [03:01] before we get there, let's talk about [03:03] some of the news that is right now. And [03:05] I think this week, even though it has [03:07] been a holiday week, some big news in [03:09] the acquisition and the aqua hire front. [03:12] Uh consolidation time. Maybe maybe [03:15] they're trying to get this in before the [03:16] end of the year. I don't know. But [03:18] Nvidia has agreed to a non-exclusive [03:20] deal to license Grock. That's Grock with [03:23] a Q, by the way. That's important. [03:25] Grock's AI chip technology. They're also [03:28] hiring away some of its top leadership, [03:31] though Grock is going to continue to [03:34] operate. Um, and you know, like I've [03:37] heard of Grock and again, this is GRQ, [03:40] not X's, G. So very important to note [03:44] that um but I'm not like intimately [03:47] familiar with them other than kind of [03:49] some of the research that I've done. [03:51] They're known for um super fast SRAMM [03:54] based AI inference chips. So this is a [03:57] this deal is very largely around [03:59] inference uh in AI their chips and their [04:02] systems that means you know low latency [04:05] high predictability for performance for [04:08] chat bots and for LLMs that sort of [04:10] stuff and um like I said it's all about [04:13] inference so it's targeting this really [04:15] competitive inference market that's [04:17] happening right now um to kind of bring [04:19] down cost increase speed and take [04:22] advantage of of that. How familiar are [04:25] you with Grock? I I wasn't at all. And I [04:28] went I'm I'm a bit of an afficionado of [04:30] watching Nate B. Jones's videos and I've [04:33] seen them. He has short versions on [04:34] YouTube. I I subscribed and so I watched [04:36] a 36-inute [04:39] explanation of this deal. Uh but Nate [04:41] was was very good and very clear about [04:43] it and emphasizing exactly what you said [04:45] Jason. Uh this is about inference. That [04:48] is to say the output of AI rather than [04:50] the training of AI. Nvidia owns the [04:53] market for training chips and and [04:55] there's a huge um shortage of what the [05:00] versus what the industry demands right [05:01] now. But this is on the other side and [05:03] and they're betting Jensen Wong is [05:06] betting that the models are trained, the [05:08] models are commodified, the models are [05:10] going to change. Yes, there'll be more [05:11] training. Yes, but the real action is [05:14] going to be in hoping that this becomes [05:15] a major business with major usage by [05:18] corporations and individuals. That's [05:20] inference. that's using the model to [05:22] give you the output that you want. And [05:25] um what Nate explained about this is [05:28] that the Gro uh chips have a fairly [05:31] small amount of memory but as part of [05:35] the processor and [snorts] so it's [05:37] extremely fast access which makes it uh [05:41] faster and cheaper to then be able to um [05:47] produce the inference output. And so he [05:50] was very complimentary of um uh Jensen [05:54] Wong uh saying this was highly strategic [05:57] that rather than saying and and I've [05:59] seen this others have said this well [06:01] rather than Jensen Wong saying well we [06:03] got to build this ourselves he said no [06:05] they're good they're out there uh and [06:08] and he went after them he went after [06:10] them an unusual deal again as you said [06:12] it's the first reports were that he [06:14] bought the company and you'd think at [06:15] $20 billion he would have bought the [06:17] company [06:18] >> but instead not was a license company [06:20] stays out there. I don't know what his [06:22] ownership stake will be in in Grock. I [06:23] didn't find that out. Uh but then he got [06:25] top two executives or thereabouts from [06:27] Grock over to [06:28] >> Ross and Madra [06:30] >> uh right over to uh John [06:31] >> Nvidia. [06:33] >> Okay. And that's Jonathan Ross who's the [06:35] founder of Grock and then Sunny Madra [06:37] who's the president. So they move over [06:40] and the former [06:41] >> That's so interesting to me from a deal [06:44] making standpoint [06:46] just this whole this whole point we find [06:48] ourselves in right now because we keep [06:50] hearing about the aqua hire without the [06:52] acquisition strategy and you know you've [06:55] got Nvidia doing it. This isn't the [06:57] first time Nvidia's done it. You've got [06:58] Microsoft, Meta, Amazon all employing [07:01] this strategy. What is Grock left with [07:03] if their top found like their founder [07:06] and their president, the people who, you [07:08] know, arguably led the company to this [07:12] major point, suddenly they're hollowed [07:15] out? What what is Grock left with? [07:18] >> Well, but again, [07:20] Nvidia didn't buy it simply licensed [07:23] technology. So, it's it's an Nvidia [07:25] interest, [07:26] >> right? It's an if Nvidia has a stake, [07:28] and again, that's the part I don't know. [07:30] if they have a stake in the equity in [07:32] Grock, which I assume they do for that [07:34] amount of money, um it's in their [07:36] interest for Grock to keep succeeding. [07:38] >> It's a very different kind of structure. [07:40] You know, Germany is somewhat like this. [07:42] This is not the same model, but they [07:44] have interlocking [07:46] corporate structures where everybody [07:47] owns a piece of everybody else [07:49] >> and they all kind of benefit if it [07:51] happens. uh this is different from that [07:53] but I think it starts to build a new [07:55] kind of structure here um where they can [07:59] do that and and um he gets the [08:01] technology he gets the view of it he [08:04] expands his business uh the other thing [08:06] I' I've read about this is that um [08:11] Google with the TPUs [08:13] >> is somewhat [sighs] I'm going to get [08:15] this completely wrong and I apologize to [08:17] everybody out there because I just don't [08:18] know enough um Google and Google Cloud [08:22] and Deep Mind models and Gemini um were [08:26] a bit more like [08:28] what Grock brings, but Google is not [08:31] selling their chips, at least not at [08:33] scale, to the world, [08:35] >> right? Yet. Yeah. [08:36] >> Right. Yet. I think they will, but [08:38] they're not yet. So, Nvidia is and it's [08:40] it rounds out Nvidia's [08:43] >> um offerings in its hosting in a way [08:47] that becomes competitive with Google and [08:49] sets them up strategically. That's [08:52] >> it makes them more integrated and in in [08:56] that sense kind of covering all of the [08:59] different aspects now covering on a [09:00] deeper level the inference aspect of [09:03] their business you know so that they can [09:05] compete with you know AMD and I guess [09:07] Cerebras is another company that you [09:10] know competes heavily in this space. Now [09:12] Nvidia, you know, is really moving to [09:15] kind of control, maybe not own because [09:18] they didn't buy Grock outright, but [09:20] control internally within their systems [09:22] all of those points of the stack. [09:25] >> Yeah. [09:25] >> Right. And as we've talked about often, [09:26] everybody thinks of Nvidia as only a [09:28] chipmaking company, but they're much [09:29] more than that. CUDA is [clears throat] [09:31] critical and to keep people on the CUDA [09:33] platform for Nvidia is critical because [09:35] that also addicts them to Nvidia chips [09:38] and Nvidia hosting. whether that occurs [09:41] at NVIDIA, whether that occurs at AWS or [09:43] anywhere else. So Jensen Wong is highly [09:46] strategic and figures it out here. So I [09:47] also read Cecil G Tamura on Facebook who [09:51] finds a lot of interesting papers and [09:54] explains some of these deals as well and [09:57] she puts it this way that in the [10:00] competitive uh landscape now I just lost [10:03] it, didn't I? Um there it is. Uh, Google [10:07] owns layers but not the broad industry [10:10] lockin. AMD is a hardware challenger but [10:13] a weaker ecosystem gravity. Open AI is a [10:17] model leader but a buyer of compute and [10:19] not an owner of it. Meanwhile, she says [10:22] NVIDIA plus Grock could unify [10:24] deterministic real-time inference, [10:27] flexible training and GPU throughput, [10:30] DPU offloads, don't ask me to explain [10:32] it. uh that keep data flowing, CPU [10:35] orchestration. [10:37] >> Um and so [10:40] >> it's really interesting just to [10:41] understand the fuller strategic scope of [10:44] these companies. [10:47] >> Well, and yeah, kind of like what we [10:48] were saying, you know, dominating the [10:50] platform as um is it Cecile? Cecile. [10:55] >> I think so. Yeah. [10:56] >> Yeah. Cecile. I wanted to say the first [10:58] time I saw it, I thought is that Sesame? [11:00] >> I [laughter] don't know. Maybe it is. [11:02] could could be either. But that's what [11:03] Cecile says here. You know, it's not [11:05] just competitive advantage, it's [11:07] platform dominance. The the whole the [11:10] whole integrated platform, you know, [11:11] super nuts, [snorts] [11:12] >> right? [11:13] >> And um yeah, that's that's interesting. [11:16] Um and by the way, you brought him up, [11:18] Nate Jones. I'm such a fan. It would be [11:21] cool to have Nate on the [11:22] >> I've tried to get a hold of him and and [11:24] I haven't got any response. He's great. [11:26] I finally so it's I just subscribed to [11:29] him and he goes on these videos at [11:32] length but he's very good at explaining. [11:34] He also loves the tools and uses them in [11:36] interesting ways. [11:37] >> Yeah. Well, he's he's got a very [11:39] [clears throat] I would say if I had to [11:41] summarize my my perspective on him, he [11:43] has a very sober perspective on how to [11:45] use these tools that isn't over [11:48] overblown. [11:50] it it he does a great job of identifying [11:53] and and flagging, you know, the the [11:55] downsides of things, but also really [11:57] smart ways to use these things. I've [11:59] learned a lot about [12:01] >> Yeah, there was there was one This is [12:03] such a side tangent, but since we're [12:04] talking about Nate Jones, I can, you [12:06] know, I'll go ahead and and mention it, [12:08] but I had set aside he had mentioned at [12:11] one point self-correction systems. Did [12:15] you did you catch his video on this? Of [12:17] course, I don't I didn't have access to [12:19] a Substack before then. You gifted me [12:21] the months and I really appreciate it [12:22] because I'm super I'm a super fan of of [12:24] his from what I've seen on YouTube. So, [12:26] I'm really going to dig into this. But [12:28] the whole idea is is that, you know, [12:31] usually when you're using these LLMs and [12:33] you're using these models, as we know, [12:36] they are going to often more times than [12:38] not give you the thing that they think [12:40] you want to see versus really [12:42] questioning and being critical on their [12:44] own output. And so his whole idea or [12:47] that he shared, I don't know if he [12:49] coined it, but is chain of verification, [12:51] which is this idea of requiring a [12:54] verification loop inside of that same [12:57] conversational turn. So essentially, I [13:01] want you to do all these things. Now, [13:04] once you've done all these things, [13:06] identify three different ways that your [13:08] analysis it might be incomplete. Once [13:11] you have those three ways, cite specific [13:14] language that confirms or refutes your [13:16] concern about those things and then [13:18] revise your findings based on all this [13:20] stuff and based on your verification. [13:22] And I've been using that for the past [13:23] month and that feels like a little cheat [13:25] code because I get such higher quality [13:28] stuff out of these systems by doing [13:30] that. So Nate's super smart. He's got [13:32] great suggestions and ideas on how to [13:34] use these things and uh highly recommend [13:36] following his work. He's awesome. Yeah, [13:38] he's a great explainer and he's he's I [13:40] would say that he's very much an [13:41] enthusiast and that comes across, but [13:44] he's not a fanboy. He's not crazy about [13:46] the stuff. He's not going off into AGI [13:48] clouds. [13:48] >> He's saying this is what just happened [13:50] and let me tell you what it can do. Let [13:51] me tell you what I did with it. Let me [13:53] explain how it works. So, I find it [13:55] useful. [13:55] >> And he always has, I mean, fresh [13:57] perspectives on everything. As you can [13:58] see on his Substack, there's a post [14:00] every single day, like literally every [14:02] single day. And they are all accompanied [14:04] with 20 to 30 minute videos. I mean, I [14:06] don't know how this dude does it. [14:08] >> Like, I look at this and I'm exhausted. [14:09] I'm like, how do you keep up this? [14:11] >> I don't have time to watch it all, let [14:12] alone [14:13] >> cadence. No, but I'm but I have a [14:15] massive amount of respect for the fact [14:17] that he can do this cuz boy, I couldn't [14:19] do this. That's just that's just [14:22] insanity. Um, anyways, [14:24] >> well, well, if if you if anybody out [14:26] there knows Nate, tell them that we're [14:27] we are fanboys of his. [14:29] >> Heck yeah. Love to talk with Nate. I [14:31] would love to pick his brain on this [14:33] show. Maybe we can make it happen. So [14:36] there is that Nvidia and its Aquaire [14:39] position. [14:40] >> We thought that was enough enough news [14:41] for the in between week. [14:43] >> Yeah, you thought that was it for [14:45] consolidation time extravaganza. [14:48] Uh we have also happening this week Meta [14:52] buying Manis which is is it manis? Manus [14:55] I I've always said manis. [14:57] >> Everything else needs to come with a [14:59] >> and a pronunciation guide. [15:00] >> Yeah, [laughter] [15:01] >> totally. or the parenthesis with the [15:03] phonetic pronunciation next to it. Um, [15:06] this is a very buzz buzzy like it's it's [15:09] been catching a lot of buzz this year. [15:11] AI agent startup Meta's deal to buy [15:15] Manis reportedly worth around $2 [15:17] billion. That's the valuation that Manis [15:20] was seeking for its next investment [15:21] round, by the way. So, Meta just kind of [15:24] said, "Here it is." And boom, we own [15:26] you. Manis, uh, a little background, [15:29] Singapore-based startup, uh, earlier [15:32] this spring, I think, is when we first [15:34] started hearing about this company, and [15:35] they were showing off their agents, [15:37] doing things like screening job [15:39] candidates, planning vacations, [15:41] analyzing portfolios, and, you know, a [15:45] lot of people were very impressed by [15:47] what it was able to do. So, now it's a [15:50] meta uh, property, [15:52] >> right? So, this company actually started [15:54] inside China. [15:56] >> [clears throat] [15:56] >> Uh and when they got the announce the [15:59] the last investment from um Benchmark, [16:03] they moved to Singapore and they're [16:06] already been controversial because some [16:07] people in in Congress are complaining [16:10] about uh Benchmark having invested in a [16:13] China based company, but now they've cut [16:15] their ties to China. I don't know what [16:17] China thinks about that. Uh, and I don't [16:19] know where the assets are and so on so [16:21] forth, but they're bought lock, stock, [16:23] and barrel by um by Meta and and the [16:27] agent thing. A couple things here. One [16:29] is that that um it brings agents are the [16:32] hot thing right now. [16:34] >> So, it brings agent uh expertise in. [16:37] Two, it's a I think primarily B2B use. [16:40] It's primarily corporate customers, [16:42] which could expand Meta's uh revenue uh [16:46] picture. uh three it's a revenue company [16:48] now so it brings in some revenue to uh [16:51] Meta's rather struggling AI business [16:54] strategy [16:55] >> yeah they've got a lot of lot of uh [16:57] recurring what was it uh zero dollars [17:00] when it launched obviously to hund00 [17:03] million [17:04] uh fastest startup oh that's right [17:06] that's what they said fastest startup to [17:07] go from $0 to hund00 million in the [17:10] world because only at eight months since [17:12] launch they crossed $und00 million in [17:14] annual recurring revenue that's pretty [17:16] significant again in a short amount of [17:17] time. [17:18] >> So, it's another case though I think [17:21] where it I mean even though we just [17:22] praised u Nvidia or I did uh for being [17:26] strategically minded and going outside [17:28] and and and acquiring by one means or [17:30] another what you don't have. [17:32] >> I I think to be a little unfair to [17:34] Zuckerberg, it makes me ask whether he [17:37] still um is struggling for a strategy [17:41] and is trying to acquire a strategy. Um, [17:44] you could argue make the same exact [17:46] argument I did for Nvidia. Oh, he's very [17:47] smart. He went out and this is hot and [17:48] he grabbed it and he used his money [17:50] because he's got it to do that. And I [17:52] think that's a perfectly legitimate [17:53] argument. But I still don't know where [17:55] Meta's [17:56] >> overall strategy is headed these days [17:59] >> with Yanco leaving with this purchase [18:01] with the prior purchase with [18:04] >> [snorts] [18:04] >> uh evidently some tourists within the [18:06] ranks. So it's going to be really [18:07] interesting to watch Meta in the next [18:09] year. Don't count Zuckerberg out ever. [18:13] Um, [18:13] >> he's got he's got enough, you know, [18:15] money in the in the bankroll to [18:17] definitely not count him out because [18:18] again, he he can buy his way into things [18:22] as needed. [laughter] [18:23] >> Yep. Exactly. [18:24] >> One of the few companies that probably [18:25] has I'm, you know, relatively unlimited [18:28] resources as far as that's concerned. If [18:30] >> he's pulling resources away from the [18:32] metaverse to AI and believes this is the [18:35] future. So, uh, you know, we'll see [18:37] where it goes. [18:38] >> Yeah, it'll Manis will keep running [18:40] independently. I think maybe you already [18:42] uh mentioned that face uh Meta plans to [18:45] integrate these agents into Facebook [18:47] into Instagram, WhatsApp on top of of [18:50] course Meta AI uh chatbot and uh yeah [18:55] it's kind of interesting in the in the [18:57] year 2025 for all that it is especially [19:00] here in the US a company like Meta [19:02] purchasing you know this this Chinese [19:04] technology company has its ties to [19:07] Tencent and Zenfund and all these things [19:10] which like You said they're shutting [19:11] down its operations in China after the [19:15] acquisition, but um interesting time for [19:18] >> Meta stock first took a dive as [19:20] acquirers often do, but today it's up [19:24] one and a third percent. So it's never [19:26] lasts long, does it? Those dips happen. [19:29] [laughter] [19:29] >> Yep. [gasps] [19:30] >> Just seems like that's the way of the [19:32] world right now. Um [19:33] >> yeah, interesting. I mean, I have heard [19:36] really I have not exper uh played around [19:38] on a deep level with Manis myself. I've [19:41] I've, you know, thrown a handful of [19:43] things at it over the last handful of [19:45] months here and there for whatever [19:46] reason, but um I have heard really good [19:48] things about it. So, yeah, be curious to [19:52] see uh what Meta does with this. [19:53] >> And to grab something away from China, [19:55] that's okay. [19:57] >> Yeah. Well, I I suppose so. So maybe [20:00] that's why maybe that's why it's okay, [20:03] you know, for for Meta as a company uh a [20:06] US company to to buy China because see [20:08] we're taking it from them, you know, in [20:11] the in the eyes of the US government [20:13] anyways. [20:15] >> Uh if that makes any sense. Um okay, so [20:18] first of all, I want to thank our our [20:20] supporters on Patreon. Y'all are [20:22] awesome. patreon.comincideshow. [20:25] We deeply appreciate your contributions [20:29] to the show and uh sticking with us. You [20:31] know, we're coming up on two years here [20:34] soon on the pod with the podcast, which [20:36] I kind of can't believe. That's kind of [20:38] crazy to me. [20:39] >> Uh how fast time flies. Uh big thanks to [20:43] patrons of the week, Matte Artac, Philip [20:45] White, Enda Malik, just a few of our [20:48] amazing patrons who support us on a [20:51] monthly basis through [20:52] patreon.com/iinsidow. [20:54] Thank you so much for doing that. Um, [20:56] and a few of you I have the uh the the [21:00] swag, the Google search swag. I think I [21:02] announced it on last week's interview [21:04] episode. Picked a couple of winners. [21:06] They're uh I got to hit the post office [21:08] today. I've got all the swag all boxed [21:09] up and ready to go. But you're going to [21:11] get them. So, you're going to get these [21:12] shirts and this hat whether you like [21:14] them or not. [laughter] [21:16] We'll see. We'll see how you feel about [21:18] that. Uh, also want to take a quick [21:21] break and thank the uh, sponsor of this [21:24] episode. Oops, it's not Jeff Jarvis. [21:26] Although, thank you for being here, [21:28] Jeff. You're awesome. [laughter] [21:30] [gasps] Thank you to uh, the sponsor of [21:32] this episode, Your 360.I. [21:37] Just absolutely love doing this this [21:39] whole process with the folks at your [21:40] 360. Jared Grock, of course, uh, had [21:44] joined me on the on the channel to take [21:46] a look at the product. And now I'm going [21:48] through the process. I think I think [21:50] everyone, including you, Jeff, have uh [21:52] have given me my review, but I haven't [21:54] gone through the portion where I check [21:57] in on the review. That's next in my [21:59] process. So, I'm going to check that out [22:02] and see what it's all about. I'm excited [22:04] to to see what everybody said about me. [22:06] But, uh here's what's really interesting [22:08] here and why this is so applicable. The [22:11] lack of career development is the number [22:13] one reason people quit a job. We're at [22:15] the end of the year, so this is an [22:16] important time to think about this [22:17] stuff. Most of us have never gotten [22:19] feedback that was really just good [22:21] enough to actually act on and to improve [22:23] our craft. Definitely true for me. Well, [22:26] your 360 does something that was, you [22:29] know, much more challenging even a few [22:31] months ago because we have the rapid [22:33] pace of development in AI. This uses [22:36] voice AI to conduct real conversations [22:38] with you and your colleagues, 15 to 20 [22:40] minutes each, drilling in on specifics, [22:43] uh, the ways that you've done well, your [22:46] growth areas, everything in between, and [22:48] then it synthesizes all of that and [22:50] walks you through the findings step by [22:52] step in a coaching conversation, which I [22:54] have as my next part of the process. And [22:56] if you're a manager, it surfaces [22:57] patterns across your team that you might [22:59] never see in a standard survey. Really [23:01] powerful stuff. A PM at Dropbox called [23:04] it the most helpful, actionable career [23:06] advice they've ever gotten. And you can [23:09] check it out for yourself, too. You can [23:10] start the year with real clarity at your [23:12] 360.ai. [23:14] Use code inside through the month of [23:16] January. You get 10% off. That's your [23:19] 360.ai code inside. And you'll get 10% [23:23] off through the month of January. And I [23:26] got to check in with you, Jeff, just [23:28] real quick because I know you did the [23:29] you did the process for me. How'd that [23:32] go? [23:32] >> So, uh, yeah, as part of part of working [23:35] at, uh, I was giving Jason a, um, a 360. [23:39] I've never been crazy about 360s. So, [23:41] all right. So, I go in and it's a [23:43] fascinating process because it's a it's [23:45] a woman as the agent, a voice, ask me [23:48] these questions, and they're and they're [23:49] clear and good questions. But a couple [23:52] times I decided that I was going to be a [23:53] jerk because I am. So, I wanted to push [23:56] the AI and see what it would do. and and [23:58] there was I forget what the question was [24:00] but but she asked and I kind of didn't [24:01] answer it and she came back I think two [24:03] more times to try to get it and I said [24:05] you've already asked me that question [24:06] three times and she well I'm sorry [24:09] you're right she said she [laughter] she [24:12] moved on u and I there's something else [24:15] there was one other point in it where I [24:17] where I pressed I think I think you know [24:19] she presses for for weaknesses and I [24:21] always hate that question if I had any [24:23] I'd say it so I I got grumpy again [24:25] >> and and she tried again and it was [24:27] absolutely fascinating to watch. So, um, [24:30] then I got email from Jared saying, "Uh, [24:33] well, your your conversation got flagged [24:35] as one as um having complaints kind of." [24:39] I I'm so embarrassed to the sponsor and [24:41] everything. And I said, "No, no, no, no. [24:42] I was just having fun." [laughter] I [24:44] just decided to see what it would do and [24:46] push it. Um, yeah. And it did very well. [24:49] It did did it did superbly. I would say [24:50] that whether or not they were sponsor. [24:52] But the other thing that was fascinating [24:53] to me about it, I said this to Jared, is [24:57] that if I were doing this with a real [24:59] person, [25:00] >> I would be more constrained. [25:04] Um, this made me feel a little freer [25:07] that there wasn't a factor of [25:09] embarrassment or that kind of stuff with [25:11] this stuff. if you just try to answer [25:12] the question [25:12] >> or saying the wrong thing like like [25:14] having an intention of what you're [25:15] saying but it coming out wrong or [25:17] whatever [25:17] >> or getting read wrong like oh I guess [25:19] he's not that fond of Jason if he didn't [25:20] say he was the most brilliant man ever [25:22] that kind of stuff [25:23] >> and so it was oddly freeing it also [25:26] freed me to be a jerk [25:29] [laughter] [25:29] >> but I was a little mean to the lady u I [25:33] love it [25:34] >> you've already asked [25:35] >> you got a stress test [25:36] >> uh they did it did it was it was she and [25:38] and she needed to do what she did I say [25:41] it um because it you know tried to get [25:44] an answer to this question and I I [25:45] honestly didn't have one um but that's [25:48] why I played it and so hats off to [25:50] Jared, hats off to uh your 360.ai. Uh, [25:53] if you do it, I think you're going to [25:55] find it's more um a better experience [25:59] than ever having done a 360 before [26:01] because they can generally be a pain. [26:02] And this is a way to dep it and to [26:05] that's not the whole pitch obviously, [26:06] but also to get to some substance to [26:09] make it useful for both the people [26:11] giving it and the person receiving the [26:13] feedback. And so I I'm I I get it now [26:15] and I think it's great. [26:16] >> Yeah. Cool. Well, I can't wait to see on [26:18] the other side. And I basically have an [26:20] email this morning that's like I think [26:21] most of my the people that I sent it out [26:23] to have completed it. So now I get the [26:25] analysis part and then I get to see the [26:28] hard truths. What am I good at and what [26:30] do I need to work on? But I'm prepared [26:32] for it. I'm all about, you know, [26:34] >> I left out one thing, Jason. I left out [26:35] one thing. [26:36] >> Oh yeah, [26:37] >> using gift as a verb. I'm sorry, man. I [26:40] can't take that. I don't like gift as a [26:43] verb. I would have added that in if I'd [26:44] heard it today. So add that in to your [26:47] feedback, man. [26:48] >> Okay. All right. [laughter] [26:51] All right. Sounds good. Your 360.ai code [26:54] inside 10% off through January. Thank [26:56] you, Jeff, for that. And uh thank you to [26:58] the folks at your 360 for sponsoring [27:00] this episode of the AI Insight [27:02] >> for your support. Much appreciated. [27:04] >> All right, we're going to take a super [27:06] quick break. Then we're going to come [27:07] back and we're going to talk about anti- [27:09] AI or at least a reaction in the [27:11] opposite direction to AI that we're [27:13] seeing right now. A few articles, very [27:14] interesting. That's coming up here in a [27:17] moment. [27:23] All right, let's let's take a look at [27:26] this anti-I movement as the author at [27:30] the New York Times. Anyways, wrote, this [27:32] is Michelle Goldberg writing for the New [27:35] York Times, the article called an anti- [27:38] AI movement is coming. Which party will [27:40] lead it? And you know there's a few [27:43] articles that we can talk about here but [27:45] she argues in this article that there is [27:48] a widespread public souring on AI that [27:52] is happening [27:53] >> and I think that you know that's that's [27:55] certainly true in some corners that's [27:59] true right yeah exactly um she says [28:02] driving this is a sense that it's less a [28:04] miracle technology and more an erosion [28:07] to art to work education mental health [28:10] privacy perception of reality all the [28:12] things. She points out that uh many of [28:14] the people that are creating and [28:15] promoting the technology have also [28:17] acknowledged that it has dangers and and [28:20] flaws, but they continue to do it [28:22] anyway. Something, you know, we've we've [28:24] definitely talked about uh on the show. [28:26] And I think her broader point is looking [28:28] at how the AI divide has an interesting [28:31] impact like on intersecting at least [28:35] here in the United States the US [28:36] political lines because you've got like [28:38] you got like President Donald Trump and [28:42] some Democratic leaders who are really [28:44] kind of supporting and courting the AI [28:46] companies uh lessening regulation that [28:48] sort of stuff. You've got Republican [28:50] leaders and independents like Bernie [28:52] Sanders kind of together in in some [28:55] capacity to push for more regulation, [28:57] moratoriums on the technology, that sort [28:59] of stuff. And it leaves voters kind of [29:03] in in a strange position, I suppose, [29:06] where they might support a candidate [29:09] deeply on a number of things, but then [29:11] really feel strongly, you know, anti or [29:14] for their stance on AI and how it's just [29:18] kind of playing with the uh the [29:20] political lines in the US to a degree. [29:23] >> Yeah. So, so Politico had a very similar [29:25] story. It's as if they'd read each [29:27] other. Uh, Americans hate AI. Which [29:30] party will benefit is their their take. [29:32] And as you mentioned, Bernie Sanders, [29:34] >> all Americans. [29:35] >> Yeah. And then Bernie Sanders is now [29:37] coming out and and arguing for a [29:39] moratorium on development, which I think [29:41] is ridiculous. Um, [29:44] and so it's fascinating to watch this, [29:47] but if we look at it in the context of [29:49] history, it's really easy to see how [29:51] we've been here before. You go back in [29:54] the 90s, uh, section 230 was a [29:56] bipartisan piece of legislation from [29:59] Republican and Democrat that defended [30:01] the future of the internet and is to [30:04] many people, including me, a critical [30:06] piece of legislation. Then you have [30:08] others from both parties in recent years [30:10] who were smashing section 230, arguing [30:13] that it should be repealed, God help us [30:14] if it were, and not really understanding [30:16] what it is, but it's just an anti-tech [30:18] view. And then I just finished reading [30:21] before the show a book by the famous um [30:24] sociologist Raymond Williams in 1975 [30:27] about television. So about as far away [30:30] from the moment when television hit the [30:33] majority of of homes in America as we [30:35] are today from the turn of the century [30:37] and uh the web taking over that. So it's [30:40] really [30:40] >> Oh okay. Wow. you know, it's a 20 30 [30:43] year old technologies. [30:45] >> And then uh television was seen in great [30:49] measure not as a cultural institution [30:52] but as a technology. Uh and then go back [30:56] another 50 years and radio was seen not [30:59] as a cultural institution but as a [31:01] technology and the reaction of many in [31:04] the public and of [31:07] um [31:09] the uh cultural institutions, critics, [31:12] academics, politicians and so on is to a [31:16] technology [31:18] >> rather than I think to what inevitably [31:20] follows once people start using it, [31:23] making it their own and it becomes [31:24] demystified as a technology. So we're in [31:26] I think a standard phase [31:29] uh where people think that oh no it's [31:32] technology and it's going to it's going [31:33] to eliminate all jobs it's going to [31:34] destroy mankind it's whatever whatever [31:37] those arguments are and I think that [31:39] that becomes um specious in the long run [31:42] but damage can be done in the short run [31:44] with bad legislation and regulation. So [31:47] this I mean Michelle Goldberg is just [31:50] assuming that uh you know an AI anti-A [31:54] movement is coming. It's going to be [31:55] there. Uh Politico declares Americans [31:59] hate AI. That's ludicrous. But that's [32:01] where we are in the the kind of digital [32:04] puritanism of our media structure. [32:08] I mean there, you know, I I can't [32:10] remember which of these articles or [32:11] maybe it was an entirely different one, [32:13] but I, you know, saw Gallup polling that [32:16] shows 80% of US adults think there [32:19] should be governmental regulation of AI [32:21] and a and like an actual slowing of its [32:24] development. There was a Pew study that [32:26] shows that uh 17% of US adults see AI [32:30] having a positive influence on the [32:32] country over the course of the next 20 [32:34] years. So then you know I suppose the [32:36] alternative you know the outside of that [32:38] is 100 minus 17 is the not positive I [32:42] guess right. Um, so I mean there there's [32:44] probably some truth to the fact that we [32:46] might be looking at least to some degree [32:49] at the next, you know, election cycle [32:52] having uh having AI as a as a pretty [32:56] involved [32:58] kind of element to how some of these [33:00] decisions are made or [33:03] >> I wonder whether where people would [33:04] really put that on their list of things [33:06] that matter. [33:07] >> Yeah, that's what I'm curious about too. [33:08] Totally. I agree. I mean, I think it's [33:10] it's the economy stupid or it's stupid's [33:13] economy as the case may be. Um, I I [33:16] think the immigration stuff is going to [33:18] matter. Uh, tariffs matter. Um, people's [33:23] health matters with what's happening [33:24] with Kennedy. Um, my politics are [33:26] showing. Sorry about that. [33:28] >> Um, and so I think those issues are [33:31] front of mind. I think media are I think [33:33] this is a year-end phenomenon. Oh, I got [33:35] to write something for the last of the [33:36] column. I have to predict what's going [33:37] on next. these these articles were [33:38] everywhere. Yeah, [33:39] >> exactly. And so I'm going to declare [33:41] this war against AI. And yeah, it could [33:43] happen. Um but I don't know that it's [33:46] politically it's going to be um a high [33:48] level issue [33:49] >> for people. Yeah. Because the because [33:51] the the so-called threat is as amorphous [33:54] as the so-called promise of AGI. [34:00] >> Well, yeah. So if if I can if you'll [34:03] indulge me I have a short little kind of [34:06] related personal journey here that I [34:08] think ties into this to [34:10] >> I gave mine a minute ago so it's t your [34:12] turn now. So, okay. So, the last the [34:16] last month or so, I've been testing out [34:18] and and kind of strategizing around a [34:21] part of my business being around podcast [34:24] consulting, some sort of podcast [34:26] education, be it consulting with people [34:28] who want to start a podcast and hey, [34:30] I've been doing it for 20 years. Let's [34:31] talk and you know, maybe I can help you. [34:33] So, I've I've been talking with people [34:35] just for free, just for fun. You know, I [34:37] kind of put it out there like, hey, if [34:38] you got a podcast, reach out, whatever. [34:40] And it's been going really well. It's [34:41] been a ton of fun. And so as part of [34:43] this, I've been thinking like, okay, [34:44] well, going into 2026, what would it be [34:47] like if part of my work was, you know, [34:50] that as an arm of my my business, you [34:52] know, a consult a podcast consulting [34:54] arm. And so I thought what I need is I [34:57] need a landing page that that is some [35:00] that I can point people to to, you know, [35:02] like a form that says, okay, what is [35:04] what is your idea? What is the challenge [35:06] that you're facing? Here's a few pieces [35:08] of information here. You know, like [35:10] who's your audience? who is definitely [35:11] not your audience, all this kind of [35:13] stuff. And the idea with that form, of [35:15] course, is fill out the form. Then, you [35:18] know, now I have the information to know [35:20] that I can help you. And they get an [35:23] email in their inbox that says, "Hey, [35:25] you know, here's here's what you sent [35:27] me. Here's what I think about it. Let's [35:29] let's fire up a a consulting [35:31] conversation." I'm making this up as I [35:33] go, by the way. Um, but I thought that [35:35] sounded like a great idea. So, I spent [35:37] two days of the the holiday break [35:39] building this out, you know, working [35:41] with like a form filling site. I can't [35:43] remember if it was type form or [35:44] something else, feeding that into [35:46] Zapier, having Zapier take the the form [35:49] and put it through like a Gemini [35:51] instance and kind of understand it a [35:53] little more and then craft an email and [35:55] then send that over to Gmail and also [35:57] put it into spreadsheets. And so, I was [35:59] like, I challenged myself to a happy [36:01] ending here. [36:03] it. No, no. I I think it heads to an [36:06] interesting ending. Okay. Nothing [36:07] nothing dramatic or or horrible comes at [36:10] the end of this. But I finally after a [36:12] couple of days built this tool and I was [36:14] proud of it. I was like, "Hey, you know, [36:15] I I set out to do this. Great. I'm going [36:18] to send out a link to, you know, a [36:19] handful of people and just have them [36:20] stress test it. Like, can you just go to [36:22] this thing and fill it out?" And what I [36:25] found kind of surprised me because I'd [36:28] say the reaction of at least half the [36:30] people, mind you, these are not all [36:32] people that are techn like in our tech [36:34] circle. Some are, but some are outside [36:35] of it. [36:36] >> And I'd say definitely on the people who [36:38] are outside of it. Um, largely. I got a [36:41] very not angry but a visceral reaction [36:45] to the fact that there was AI involved [36:47] at all. [36:48] >> Wow. [36:48] >> That that like, oh yeah, your form was [36:51] great. But then I got the email and the [36:53] email felt very inauthentic. You know, [36:55] they're like, "We know your brand. We [36:57] follow you. We're fans." And the email [36:59] doesn't doesn't sound like you. Um that, [37:03] you know, the AI made certain [37:05] assumptions that collided with my [37:07] personal life in in coincidental but [37:10] alarming ways. And it was just an [37:13] interesting thing that I bring up only [37:16] now only to say that like I think there [37:19] is this kind of [37:21] uh reaction in the in the opposite [37:24] direction that I'm noticing with people [37:25] in my life that aren't in technology [37:28] that aren't steeped in this stuff, [37:30] >> right? [37:30] >> Where where people do, you know, and [37:33] it's totally anecdotal, I I realize, but [37:35] people who don't follow this stuff super [37:37] closely are starting some of them are [37:40] starting to feel kind of like an [37:41] immediate negative reaction to AI as a [37:45] thing. And that could be because of news [37:47] stories like these. That could be for [37:48] any number of reasons, but it's there. [37:51] And I really wonder with that being [37:53] there and kind of the reaction that I [37:55] saw from this thing, what does that mean [37:58] when we head into, you know, the next [38:00] couple of years where the these tools [38:02] build out even further where government [38:04] makes the decisions that it's going to [38:06] make for right or for wrong to either [38:08] bolster it or or not. I don't know. I'm [38:10] very curious about this. I'm very [38:12] curious to see what kind of impact it [38:13] has. [38:13] >> Yeah. Yet, I wonder how many of those [38:14] same people who were suspicious of AI [38:16] would say the same thing about Google [38:18] and Meta and Amazon and stuff. Is it is [38:20] it an anti- uh tech company? Is it anti- [38:23] big company, anti- capitalism? I'm not [38:24] trying to read too much into it, but I [38:26] think there's a lot of that that comes [38:27] in. Or is it specifically AI that has [38:30] cooties? I'd be very interested. It'd be [38:32] very interesting to do a poll of them. I [38:33] mean, almost go back to them and say, [38:35] uh, okay, I hear you. I got it. Um uh I [38:41] don't know how you'd arrange that that [38:43] question, but it'd be interesting to [38:44] interview them to see where their [38:45] hostility really lies because I think [38:47] that'll say a lot. [38:48] >> Yeah. Yeah. It's it's a curiosity that I [38:51] have though. Um yeah. And uh Daniel in [38:54] um in chat said, "Did you feed [snorts] [38:56] the AI some examples of your writing so [38:58] it could possibly write in your own [38:59] voice?" Yeah, I absolutely did. Um I [39:02] think I think though there are just [39:04] certain tells. AI has tells and the more [39:07] we use these things no matter how good [39:09] we get them they're still there whether [39:12] we recognize them or not they give us a [39:14] certain feeling you know and I don't [39:16] know it's just interesting now it has me [39:18] rethinking it like maybe I don't want [39:19] this thing to be too smart [39:22] >> right it's because they're they're it's [39:23] you trust yeah so I just got caught by [39:26] slop um I'm I'm um affiliated with the [39:29] Alan Alda Center for Communicating [39:31] Science at Stonybrook uh really [39:33] wonderful institution and Through that, [39:35] I've met Al had the honor of meeting [39:36] Alan Alda and speaking with him a couple [39:38] times and just an amazing amazing man. [39:40] And I'm going through my Facebook and [39:42] there's a journalist, a journalism [39:43] educator I really respect who put this [39:45] story up of how Alan Alda helped one of [39:48] the cast members of um MASH in the day. [39:52] And so I shared it with my boss at [39:56] Stonyiebrook because she's good friends [39:57] with Alan Alda. And she said, "Oh, what [39:59] a wonderful story." And about an hour [40:01] later, I had to come back and say, "I [40:02] got slopped. Uh, somebody showed how it [40:05] was slop. And when you think about it in [40:07] hindsight, it was too perfect. It was [40:09] too this. It was too that. I I Yep. Yep. [40:12] >> You got slopped. [40:14] >> I got slopped. [laughter] [40:14] >> I love I love that term. [40:16] >> I got But here's a better story. So, I'm [40:18] not going to say what it's about because [40:19] I want to keep this [40:21] quiet. Um, and you may laugh at me for [40:24] this, but I have an idea for a [40:26] screenplay. [40:28] >> Okay? [40:28] >> And I have no idea how to write a [40:30] treatment. No idea whatsoever. And it's [40:32] probably insane, but it was a same [40:35] thing, holiday week. Oh, what the hell? [40:36] I'll spend a day and a half doing this. [40:38] Uh, and um, uh, I started by writing [40:42] about a four-page just narrative of it [40:45] and then I fed it to notebook LM and I [40:47] said, can you help me write a a [40:49] treatment and again it has all this [40:51] formatting thing? Act one, act two, act [40:53] three, the this, the that, there's all [40:55] this way to do it, right? And I tried [40:56] the same thing with JGPT and Gemini and [40:59] Anthropic, I think. And the formatting [41:00] was the same in all of them. [41:02] >> I think what they did was pretty much [41:04] crap. Uh it was still my story. I knew [41:06] what the story was, so I wrote it. But I [41:09] could have read uh anywhere online. I'm [41:12] sure I could find how to write a a [41:14] treatment. [41:15] >> But then it's antiseptic and other by [41:20] using my own material and making me [41:22] critical. Well, I can do better than [41:23] that. Well, then okay, Jarvis, do better [41:25] than that. Okay, I will. Right. It was [41:27] you more useful as an educational [41:29] exercise for me [41:31] >> um because it was it was it was relevant [41:33] to me. It I understood what was wrong [41:36] with it. I could go with it and it was [41:38] really kind of an it was my first best [41:41] collaborative effort with AI. So in the [41:44] end it was me. I threw out everything [41:46] that it said but I learned from it [41:48] nonetheless. It was helpful. [41:50] >> Yeah. And did it did it lead you to make [41:52] different choices in what you chose to [41:54] do from a creative standpoint [41:56] afterwards? Maybe it you know [41:59] may maybe it leads you to to oh I hadn't [42:01] thought about that particular thing or [42:03] whatever. So maybe it influences you in [42:05] ways other than hey I wrote your [42:06] treatment for you. Do you like it? [42:10] You know the problem I have with it? So [42:12] I wrote I wrote to my agent. I sent it [42:14] to my agent. Uh I haven't heard back [42:15] from her. Uh the the subject line of the [42:17] email was you'll think me insane dot dot [42:20] dot. Um uh so I have no idea what she'll [42:23] say. What bothers me about the AI was [42:26] the sycopancy. [42:27] >> Wow, what a great story, Jeff. This is [42:29] really filable. Well, I don't know [42:32] whether that's true. It's a freaking [42:34] machine, [42:35] >> right? And I don't want to I don't want [42:36] to take any of that in. [42:39] >> Um but it's fascinating to work with [42:41] this stuff. And again, I wrote a [42:42] syllabus for Stonyiebrook uh for a [42:45] course in AI and creativity, which I [42:46] didn't teach. Two two great people [42:48] didn't teach it this term, and it filled [42:49] up immediately with 300 students. And [42:51] next term, another person I really [42:52] respect theorists teaching, it's 500 [42:54] students. And and yeah, there may be [42:57] people out here who hate II or or want [42:59] to growl about it or want to give it up [43:00] or all that, but it ain't going away, [43:02] folks. Yeah. [43:05] Um Yeah. Super interesting. Like gosh, [43:09] the show could be about these these kind [43:11] of stories because I think I think you [43:14] and I having these experiences is [43:16] indicative of what a lot of people are [43:18] experiencing as we work with these [43:19] things more. I mean, my wife is is [43:22] working more with these. I got her I got [43:23] her a new laptop for Christmas. I was [43:25] very proud because anytime I had to use [43:26] her her laptop for her business, I [43:28] wanted as I told my daughter, I just [43:30] want to jump off the roof. It's so [43:31] miserable and frustrating. Everything [43:32] takes a million hours just to do. So, [43:35] got her a new laptop because I've been [43:37] also been kind of showing her what I've [43:39] been learning about working with these [43:40] systems for her business and everything. [43:42] And she told me the other day, you know, [43:44] she was sitting there, she was like, you [43:45] know, these tools are fine and good and [43:47] like it's it's it's interesting and [43:49] magical to see what they can do [43:50] sometimes. She's like, but the more I [43:52] work with them, the more I realize how [43:54] imperfect they are. [laughter] [43:56] >> I was like, yeah, spot on. Like, yes, [43:59] >> they seem so magical and so, oh my [44:01] goodness, it can do everything for me. [44:02] Now, when you first start, the more you [44:05] get accustomed and work with them, the [44:06] more you really push up against the [44:09] obvious limitations, unless you fall [44:11] down that that psychicopantic route of, [44:14] oh, I am really great. I this this is [44:17] amazing. This is the perfect way to [44:18] phrase things or whatever. And then you [44:20] you know and and I think I kind of faced [44:23] that a little bit with my tool and [44:25] thinking like oh I built this great tool [44:27] but not thinking about like what is the [44:30] what is the message that I'm conveying [44:32] on the other end when they get that [44:34] final email and it's definitely not [44:36] authenticity which is a core value of [44:38] mine and so okay then I have to scale [44:41] back and I have to think about this [44:43] differently because that doesn't work [44:44] for me. [44:45] >> Yeah. [44:46] >> Yeah. Interesting stuff. Uh, I love that [44:49] tangent. That was fun. Um, before we get [44:52] to the archive showdown, just real [44:54] quick, because everybody was talking [44:55] about this uh this story, OpenAI hiring [44:59] ahead of preparedness, a very highly [45:03] compensated role uh if anyone wants to [45:06] put themselves up to uh this role, 55k, [45:10] thousand plus equity. [snorts] And uh [45:13] basically the whole idea here is this [45:15] unfortunate soul has to monitor and [45:18] contain risks uh from the company's [45:21] models as they increase in capability [45:23] and everything. They have to address [45:25] potential harms to mental health, [45:27] autonomous cyber attacks, misuse in [45:30] biology, misuse in weaponry, all of [45:32] these things. And uh even Altman admits, [45:35] quote, "This will be a stressful job and [45:37] you'll jump into the deep end pretty [45:40] much immediately." So Jeff, are you up [45:42] for the challenge? [45:43] >> Uh, cuz I'd walk in and say, "This job [45:45] is BS. There's no AGI. There's no end in [45:49] humanity. I'm going to sit back and take [45:53] my paycheck." [45:54] >> You're going to hang out on the roof all [45:56] day and [45:56] >> Yep. [45:57] >> Yep. Eat your lunch. [45:59] >> Uhhuh. [laughter] [46:00] Free lunch probably. [46:02] >> Probably. Oh, yeah. Probably a pretty [46:03] good free lunch, actually. [46:05] >> Yeah, I bet. [46:05] >> Um, yeah. So, anyways, I thought that [46:08] that's interesting. Definitely a lot of [46:10] people thought that interesting. Do they [46:12] think it's interesting because it's [46:13] highly compensated or because it's open [46:14] AI kind of it kind of [46:16] >> I think it's open AI trying to argue [46:18] we're going to kill the world. So you're [46:19] the you're the person who's going to [46:20] save the world from us which is so [46:22] weird. [46:23] >> Yeah. [46:23] >> That's their that's their marketing. [46:25] >> Are they are they going down the [46:26] anthropic route with this? [46:28] >> Oh yeah like [46:29] >> Well, don't forget that anthropic people [46:30] came from OpenAI and so they they're of [46:32] a similar uh DNA. [46:34] >> That's true. Cut from cut from the same [46:36] cloth [46:37] >> as they all seem to be. [46:39] All right. Well, why dance around it any [46:42] further without some music to dance? [46:45] >> As you might. [46:46] >> It's time for Jeff's archive showdown. [46:49] >> Jeff's archive showown. [46:51] >> Down and out. Okay. What you got? [46:54] >> Three quick papers uh here from my uh [46:57] two weeks worth of trolling of [46:58] archive.org the pre-print server. Uh the [47:01] first one interested me is sell it [47:02] before you make it. Revolutionizing [47:04] e-commerce with personalized AI [47:05] generated items. Importantly, this comes [47:07] from China. So, this is not going to be [47:10] anything unusual to somebody who knows [47:12] um Kickstarter. Uh uh but what they're [47:16] doing is they're using AI to design u [47:20] products without having to prototype [47:22] them and then they can put them up and [47:24] they can see do people want this and if [47:26] they want it then we'll make an offer [47:28] for it and then we'll take the orders [47:30] and then we'll make it. So it's it's an [47:33] extension of the Kickstarter model where [47:34] you use your customer's capital to build [47:37] the thing. But in a world of uh high [47:40] turnaround fashion, uh this will be even [47:42] more high turnaround. And it's not hard [47:44] to see. I think this is not really part [47:47] of the paper, but that I could see in [47:49] it. But I I does the tool need to be in [47:52] the control of just the company. [47:54] if if I designed [47:57] uh a jacket and I convinced enough [48:00] people to say, "Do you like my jacket?" [48:02] And uh what if uh they could discover [48:04] that um yeah, they'll make it for me and [48:07] they'll sell it to me and they'll give [48:09] me a cut of it because I designed it. Do [48:11] can I use AI to do that or can I get a a [48:13] a single um uh purely personalized item [48:18] because AI can do it? So it's just AI is [48:19] gener is is now moving in to this um [48:23] sequence of product development. And [48:25] I've long argued I argued many [48:27] interesting [48:27] >> that you've got to move the customer up [48:29] much earlier in the design chain rather [48:31] than designing the whole product in [48:33] secret and saying here don't you love it [48:34] and you spend a lot of money to prove [48:36] that you love it. The more that it comes [48:38] from the need and desire of people [48:40] themselves the better. So, [48:43] >> man, you know that what that has me [48:44] thinking is um we've talked um in the [48:49] past about [48:50] I I don't know if it's a possible [48:52] future. I don't know if it's an [48:53] inevitable future, but the future where [48:56] you know maybe things exist like maybe [48:58] maybe movies exist, but there's also [49:00] like an a tangent where people the [49:04] systems are good enough to where I have [49:05] an idea and I say, you know, the movie I [49:07] want to watch right now is a movie about [49:08] this guy that blah blah blah. I spell [49:10] out the the basic plot line and boop, it [49:13] creates the movie for me and it's good [49:14] enough that it's entertaining and blah [49:15] blah blah. I wonder if that could happen [49:17] with products, too, where it's like, you [49:19] know what? What I really want is a [49:20] sweatshirt that does this. Just make it [49:22] and produce it and send it. And God, I [49:25] could I could see that happening. Like, [49:27] I could absolutely see that happening [49:28] with systems like this. [49:30] >> So, uh, Google announced some things [49:32] that they that they changed and one of [49:34] them is um that they said, let me see if [49:37] this if I can get this to happen. Um, [49:42] tell me a story about an entrepreneur [49:46] who starts a company and becomes rich. [49:52] >> In Seattle, Elias Thorne didn't feel [49:54] like a future billionaire. He felt like [49:56] a man who was tired of losing his keys. [49:58] Elias was a software engineer with a [50:00] frustrating trait. He was obsessed with [50:02] micro inefficiencies. He noticed that [50:04] people spent an average of six months. [50:06] >> Okay, that's enough. Yeah. [50:07] >> But one of the things is Google is [50:09] Gemini is now saying, "We'll tell you a [50:11] story." [50:12] >> Yeah. [50:12] >> Exactly what you just said. Tell it the [50:14] story you want to be told. [50:15] >> Uhhuh. [50:17] >> Yeah. And I mean, you know, is it a good [50:19] story? That that remains to be seen. [50:21] Probably not. But [50:23] >> I mean, long enough timeline, you know, [50:26] enough development, enough training [50:28] around what is good versus not. I guess [50:30] that's the real challenge. But [50:32] >> yeah, I could see that. And and so, you [50:34] know, are we are we looking at a future [50:36] where it's like that the clothes that [50:37] you wear potentially are the clothes [50:39] that you just had an idea you wanted and [50:40] you didn't have to like spend a million [50:42] hours trying to find the thing that [50:43] matches. You just say, "I need a I need [50:45] a black turtleneck that fits a tall [50:47] size, blah blah blah, and has this logo [50:50] on the back or I don't I don't know what [50:51] it looks like." And then the AI just [50:53] goes, "Boop, here you go. We'll send it [50:54] out." [50:55] >> Yeah. [50:56] >> Six hour shipping. [50:57] >> Second second paper. Yes. Yes. Probably. [51:00] Yeah. Instant manufacturing a two door [51:02] houses down from you. The factory drives [51:04] up to your your driveway and drops it [51:06] off. [51:07] >> Second is fascinating especially what [51:08] you talked about with your wife's [51:09] attitude toward AI. What human horse [51:13] interactions may teach us about [51:14] effective human AI uh interactions. [51:18] >> Okay. [51:19] >> This from Muhammad Hussein Jihari and [51:21] Stanley Ahalt at UNC Chapel Hill. And [51:27] their argument is is simple and [51:28] straightforward. But but charming for [51:30] what it is, saying that that the way [51:33] that a rider and I'm not a horse rider. [51:34] Are you? Do you ride horses? [51:36] >> No, I don't. No, I don't. No, [51:37] >> but [laughter] you know, I've got [51:39] friends who who are I live in horse [51:40] country. Um, and there's friends who [51:43] are, but there's a certain magic that [51:44] occurs of communication that the horse [51:47] is the one doing what you're doing, but [51:49] they they have a shared and harmonious [51:51] decision- making. The horse is deciding, [51:54] you're deciding. You're coming together [51:55] to do this stuff. And so it's an [51:56] argument for a model around what to [51:59] think of as AI, which I think is is [52:01] good. It's not that it takes over the [52:02] world. It's that we work in harmony [52:05] >> in tandem. Yeah. Yeah. Right. [52:07] >> Yes. And then the third paper from folks [52:10] at uh institute polar technique in Paris [52:13] and um uh a co tech pillow technique uh [52:17] in Paris uh fake parts a new family of [52:21] AI generated deep fakes. What's [52:24] interesting about this is this is not a [52:26] full deep fake. This is the discovery of [52:29] the ability of AI to fake just part of [52:33] an image or part of a video. [52:35] >> So imagine if you would um you can [52:38] change the face, right? We know it can [52:40] do that in photos. [52:42] >> So the next phase of deep fakes, it'll [52:44] be harder to tell because 95% of the [52:47] photo is real. [gasps] [52:48] >> Wow. Okay. [52:49] >> Just this part is faked. [52:53] So, okay, that's that's interesting. [52:56] There there's probably a term for this [52:58] that I'm just not aware of. But when I [53:00] think of like nano banana, let's say, [53:02] and nanobanana has gotten really good at [53:05] giving it a source image and saying this [53:08] but here, or whatever, you know, and [53:11] retaining the the the qualities or this [53:13] image, but change my the hair to red. [53:16] >> My understanding about it is that it's [53:18] not actually only changing the hair to [53:21] red. It's doing a a whole pass over the [53:24] thing. It's just getting really good at [53:26] replicating the original thing. I wonder [53:28] how that applies to here because in that [53:30] case, is it really possible with current [53:33] systems to only change a portion of the [53:36] thing and not the entire thing just make [53:38] it adhere really closely if that makes [53:40] sense. [53:40] >> They list six uh no eight uh possible [53:43] some of which I don't understand. um uh [53:45] face swapping in painting which we've [53:47] seen that you know you you fill in part [53:49] of the was that interpolation you extend [53:52] something out painting you extend the [53:54] image a style change uh those are the [53:57] six ways that they define it can be done [54:00] >> and they're you know they're looking at [54:01] it in a in a way that I can't fully [54:03] understand because these papers are [54:04] highly technical but interesting trend [54:06] coming [54:07] >> yeah and I guess yeah that's true that [54:10] that kind of um that kind of answers my [54:12] question I mean especially when you Talk [54:14] about face swapping. I mean, this is a [54:15] technology that we've had for a number [54:17] of years now. AI has inherently it's [54:20] made it higher quality and and more [54:22] effective. And in that application, [54:26] we're not seeing the system completely [54:28] generating the entire image and just [54:30] swapping the face. So, [54:31] >> that would that would seem to be the [54:32] answer. Yeah. Is that yes, we do have [54:34] the the ability to do this. So, [snorts] [54:36] interesting. H I mean that is that does [54:39] make you know detecting a lot more [54:41] difficult when 95% of the image is is [54:44] fine and maybe it's that one really [54:46] important smaller part [54:47] >> and all you want to change is whether [54:49] they it was a $1 tip or a thousand [54:51] dollar bribe, you know. Yeah. It's all [54:53] you're doing is changing one little [54:55] aspect of of the image. [54:56] >> Yeah. Interesting stuff. [54:59] All right. Well, uh this show is [55:02] fascinating me to no end. Hopefully it [55:04] is to you, too. And if you like it, [55:06] leave us a review. I'll just say that. [55:08] Wherever you can leave a review or [55:09] comment on the videos or whatever, [55:10] interact with the with AI inside [55:12] wherever you happen to be watching or [55:14] listening. We really appreciate it. [55:15] Going to take a super quick break. Going [55:17] to come back with a few minutes to take [55:19] a look at the year that was 2025 and [55:21] maybe a look forward to 2026. That's [55:24] coming up in a moment. [55:29] All right. 2025 and review. And I think [55:31] this is inspired. you had you had put in [55:33] a link to Andre uh Kaparthy's post2025m [55:39] >> quite volume review [55:40] >> and this was for him this is extremely [55:42] short [55:44] >> extremely short [55:45] >> yes [55:46] >> okay so okay I wonder I wonder why hey [55:49] you know what we all have a lot of [55:50] things going on right now so it's okay [55:53] >> he's a little busy um but yeah so so [55:55] this is interesting so some of this [55:57] stuff I had to kind of like noodle on [55:58] and think think about I don't know how [56:00] deeply we want to go into this But uh [56:02] reinforcement learning from verifiable [56:04] rewards. So this is basically like a [56:06] training a step of training where models [56:09] are learning their reasoning um in a in [56:12] a more enhanced way because they're [56:14] chasing rewards that can get checked [56:16] automatically is I guess one of his [56:19] reflections from the year. He's he's [56:21] probably I mean obviously clearly he's [56:23] looking at this stuff on a deeper [56:24] technical level than probably I do. [56:28] >> I I'm not going to speak for you. Um, [56:31] oh, [laughter] [56:33] >> jagged intelligence, which at first I [56:35] was like, well, what the heck does that, [56:36] but it actually makes a lot of sense. A [56:38] and and kind of ties in with what we [56:39] were looking we were talking about a [56:41] little bit earlier. AI can look really [56:43] brilliant when you're talking about a [56:45] few narrow areas, but then it falls [56:46] apart on the tasks that that surround [56:49] that. Um, tasks that seem like they [56:51] should be very easy. And this is still a [56:53] challenge with AI as far as what we're [56:55] seeing. Um, any I mean there's there's a [56:58] few others here. any stand out for you [57:00] in in particular? [57:01] >> Well, I think a cursor is the hot thing [57:03] that people are going for uh this year. [57:05] I I think my son is using it as a [57:07] developer and I think it's imp [57:08] impressing people alongside clawed code [57:11] has really been the winner of the year [57:13] when it comes to coders alongside vibe [57:15] coding. [57:16] >> So in the coding world, you know, if if [57:18] if if these early stages of generative [57:20] AI had stuck to coding, I think it would [57:23] have absolutely blown the minds of a [57:26] small but powerful part of the country. [57:28] But it wouldn't have been the huge [57:30] phenomenon of business that it that it [57:32] has been. [57:33] >> Yeah. Yeah. [57:34] >> That's where it's most successful I [57:35] think. [57:37] >> Yeah. And then um it calls out Nano [57:39] Banana and LG UI which I you know I Yes. [57:42] Nano Banana is really impressive as [57:44] we've talked about. Gemini and some of [57:46] their integrated Google's integrated [57:48] kind of uh graphical hooks that are [57:51] coming into the model are is getting [57:53] really impressive and really good and [57:54] kind of um making it a more friendly [57:58] familiar kind of uh way to do this sort [58:01] of work. And so that's going to I think [58:03] over time broaden out and uh bring more [58:06] people in. So yeah, kind of a kind of a [58:08] good walkthrough. What are a few things [58:09] that that kind of stood out for you uh [58:11] for this year? I mean obviously there [58:13] are bazillion things we talked about but [58:14] I'm curious to know. [58:16] >> So uh each of us has three things that [58:18] we're we'll talk about from the year. Um [58:20] my first is the split between LLM [58:23] believers and role model believers [58:25] >> and I think we we talked about that a [58:27] lot of the show and uh you Google is [58:29] still going strong um in the LLM world [58:34] and they believe that it's going and and [58:35] when I went to this event with a a deep [58:37] mind person and Yan Lun they ended up in [58:40] a debate. So, interestingly, um, uh, [58:44] there's a video of Yan online in which [58:46] he says there's no such thing as general [58:47] intelligence. Human intelligence is [58:49] super specialized for the physical [58:51] world. This is playing into exactly what [58:53] he's doing in the future. He's working [58:54] on um world models. Feel Lee is working [58:58] on world models. Jensen Wong has talked [59:00] about world models. Uh, so in comes [59:02] Demis Hassabis who then says, "Yan is [59:06] just plain incorrect here. He's [59:08] confusing general intelligence with [59:10] universal intelligence. Brains are this [59:12] exquisite and complex phenomenon that [59:14] can take on general knowledge. And [59:16] that's the argument on that side is that [59:17] we've seen the performance so far. If we [59:20] just scale it, it'll do anything. And [59:22] then Yan comes back and says I think [59:24] there's a disagreement largely one of [59:26] vocabulary that he objects to this idea [59:28] of general intelligence. And he says [59:30] that we are all specialized in our [59:32] intelligence. There are certain things [59:34] we can do really well. We can figure out [59:35] the physics of a ball coming our way, [59:37] but as he said in the video, we're [59:39] crappy generally at chess. The machine [59:42] can be better at chess. We are [59:44] specialized in our knowledge. So this is [59:45] the essence of an argument uh that's [59:47] going on in mathematical terms and in [59:50] scale terms going on. So that's one. [59:52] >> Two is uh the second thing for the year [59:55] I think my conclusion is Google's the [59:56] new Google [59:58] >> and we a lot of people were a [59:59] >> big year for Google. [60:00] >> Yes. people were declaring Google dead. [60:02] Open AAI's taken over. Uh search is [60:04] dead. Uh no, Google's done a superb job [60:07] and I think it's it's it's on top. And [60:09] interestingly, I just threw this in at [60:11] the last minute. Uh you'll remember that [60:13] when Google put up the um this the AI as [60:16] part of the search results and somebody [60:18] from Business Insider went in and got it [60:20] to to say, "How do you get to keep the [60:22] the cheese on the pizza?" And Google [60:24] said you mix in glue, which everybody [60:27] had to make fun of. Indeed. And the and [60:29] Katy Neopoulos at Business Insider, in [60:32] fact, made pizza with glue and wrote it [60:35] and ate it. [60:36] >> Oh, [60:36] >> anything for a story, man. Anything for [60:38] a story. But now, big credit. She came [60:42] back at the year end and ate Crow [60:44] instead [60:45] >> and said that Google AIO reviews had the [60:47] last laugh this year, that they've [60:49] really been good. [60:51] >> I guess Google has the last laugh now. I [60:53] love that. [laughter] [60:53] >> Uh, they do indeed. uh and and she ate [60:57] crow without glue. [60:58] >> Um and my third is US versus China I [61:01] think this year uh because of deepsek uh [61:04] I think that's a huge part of the story [61:06] and it's going to have policy political [61:10] strategic business implications going [61:12] forward what were your thoughts [61:14] >> 100% 100%. Yeah. And by the way, like I [61:17] don't know that there's probably plenty [61:19] more, but when I was thinking about it, [61:20] I was like I was like, h here's a couple [61:22] of things that come to mind. Um, and [61:24] kind of what I was talking about [61:25] earlier, just a general kind of thing [61:27] that has me really interested right now [61:29] is this this like confused state that I [61:32] know people as more and more people are [61:34] interacting themselves with AI that I [61:36] think a lot of people are feeling right [61:37] now is, you know, both techies and non- [61:39] techies seem to, you know, they're using [61:42] the tools even though they have they [61:45] hate to admit it. [laughter] there's [61:46] like a conflicted relationship that I [61:48] think a lot of people have a lovehate [61:50] relationship kind of confusion around [61:53] around these tools because yeah and I [61:55] and I and the only reason that this is [61:57] really top of mind is kind of like the [61:58] story that I told earlier and some of [62:00] the conversations I've had with people [62:01] outside of tech circles lots of people [62:04] are using these tools and some people [62:06] are are being really um [62:09] >> I don't know some people are using them [62:10] but then don't want other people to know [62:12] that they're using them. It's like [62:13] they're ashamed to be using them. It's a [62:15] weird time for this technology, you [62:17] know, and and for people to know what is [62:19] it okay? Is it okay that I use this [62:21] technology even though everybody's using [62:23] it and everybody's pretending like they [62:24] aren't. So, I don't know. I think that [62:26] that's been um kind of an interesting [62:27] thing to watch. um just the fact that [62:31] like this was a big year for agents and [62:34] yet man do they need a lot of work like [62:36] that and and I guess agents is a broad [62:38] term but when I think of you know from a [62:41] consumer perspective which is largely my [62:43] own perspective um any of the times that [62:45] I've really kind of leaned into agents [62:47] to test out something or try something [62:50] or oh wow that capability is here we can [62:52] finally do that rarely am I ever [62:55] presented with oh it did it perfectly [62:56] yay I'm going to do that again next time [62:58] you But it's [laughter] [62:59] it's it's kind of the it's still early [63:02] days for agents as much as this seemed [63:04] like the year of agents and um again I [63:07] know that's a loose terminology and and [63:10] there are a lot of ways in which you [63:11] know people are have become really good [63:14] at programming agents on a deeper level [63:16] and getting them to do cool things. Yes, [63:18] that's possible. But I think by and [63:20] large they need a lot more work before [63:22] the masses, before more people gain [63:25] their trust and actually decide like, [63:27] okay, yeah, I'm I'm going to rely on [63:29] this for buying or whatever the case may [63:31] be. And then um did we even talk about [63:35] the AI for companionship? I think maybe [63:37] that was a story that was in there [63:39] earlier that maybe got cut. But just as [63:41] a general theme, this has been a real [63:43] interesting year when it comes to the [63:45] chat bots and their development and how [63:48] that is working its way into general [63:51] users from a companionship perspective. [63:54] Lots of resistance um notable adoption [63:58] but plenty of resistance. And I think [64:00] there's little acknowledgement in that [64:03] as we talk about little acknowledgement [64:05] around the reasons that are driving [64:07] people to do that in the first place. [64:09] And there was an HBR article that you uh [64:13] I can't remember if you put this in. No, [64:14] actually I found this on Harvard uh [64:16] business review that has a little chart [64:18] here where they looked at the top 10 [64:20] geni genai use cases in 2024 compared to [64:24] 2025. 2024 the top use case was [64:27] generating ideas. that dropped to number [64:30] six in 2025. The top of what they're [64:32] saying is therapy and companionship. And [64:35] I think that's that's an interesting uh [64:38] kind of place that we find ourselves [64:39] with these chat bots as they continue to [64:41] proliferate. [64:41] >> Yeah. We we had two headlines in here. [64:44] Uh one was from the Wall Street Journal [64:47] with doc AI chat bots linked to [64:50] psychosis say doctors. [64:52] >> Yeah. [64:52] >> Uh a little much I think. Um uh and [64:55] basically what the story says is people [64:56] already had these problems and then they [64:58] were using the AI and the AI validated [65:00] what they were saying. At the same time, [65:03] the Guardian, which tends to be, and I [65:04] love the Guardian, but they tend to be [65:06] puritanical about technology these days, [65:09] had a story saying, could AI [65:11] relationships actually be good for us? [65:13] So, it's like is red wine going to kill [65:14] you or save you? Depends on [laughter] [65:17] you last. [65:19] >> That's totally it. Yes. [gasps] Uh [65:22] there's there's so many more. I'm sure [65:23] y'all are are screaming a million things [65:25] at us, but um what about 266 real quick? [65:28] >> How about next year, Jason? [65:29] >> Yeah. Okay. I'm just going to throw [65:31] these out here real quick because I know [65:32] we are we're going a little long at this [65:33] point. Um I think I think this next year [65:36] is going there's going to be a little [65:37] bit of a reckoning because I'm kind of [65:40] seeing it a little myself as well. Uh [65:42] reckoning around how productive we [65:44] actually are when we use these tools. [65:46] Like the more the more our eyes open as [65:49] at least I can speak for myself, the [65:51] more my eyes open to the possibilities, [65:52] the more I'm like, "Oh, well, I want to [65:54] do more using these tools because that's [65:55] going to enable me to do so much more." [65:57] But then I realize on the other end that [65:59] it just it creates a different type of [66:01] work that I have to then do to make sure [66:04] that the time that I saved over here is [66:06] okay. And I kind of think in some ways [66:09] it's kind of a net equal to a certain [66:11] degree. And I wonder how how more people [66:13] are going to come to terms with that. [66:15] you know, this was magical at first. Now [66:17] I'm starting to really get a clear [66:19] picture. Is it actually making me more [66:20] productive or is it just making my my [66:22] work different? And I think people might [66:25] see that a little bit more. Uh AI [66:27] shopping moving beyond the demo. Again, [66:29] that could that has a big trust [66:31] component, but I think we're early on [66:32] the AI shopping thing. And if we know [66:34] anything about the internet and how it [66:36] all works, that's that's pretty [66:38] critical. that that moves the industry [66:39] in a lot of ways when they figure out [66:41] these new ways to buy things and to sell [66:43] things and AI is squarely in the in the [66:46] crosshairs there. I remember when I [66:49] worked for Kany ask way back in the day [66:51] u there was the belief that people were [66:53] not going to use the very belief was [66:55] people aren't going to use the internet [66:56] to buy things they're not going to trust [66:58] the internet obviously I remember that [66:59] then they weren't going to buy expensive [67:01] things and we started a style.com store [67:03] with Neiman Marcus with very expensive [67:05] things and and so these things change [67:08] over time I think you're right the [67:09] shopping stuff will get real I just have [67:11] you used Roffus at Amazon [67:14] uh I have that's where it like suggests [67:17] certain things. [67:17] >> It comes up to the sides, right? And [67:19] I've been ignoring it. Um, this is more [67:21] than you want to know about me. I've [67:22] seen it. It's cold here. So, I needed a [67:24] new robe that was warmer. And I went to [67:25] the store and they and they were like, [67:27] "Creepy feeling." And I and I and I [67:30] looked at Amazon to look at robes and so [67:33] Roffus came up and I said, "I don't want [67:35] uh a a a robe that feels creepy and too [67:38] soft and all this. I want something that [67:39] feels like a towel." Said, "Oh, you [67:41] don't want this. You want that." And it [67:43] made recommendations and I thought, "Oh, [67:44] okay. I I get this now." And I got [67:47] something in the in the shopping cart as [67:48] we speak. [67:50] >> Okay. Well, I'll be curious to know if [67:51] it feels like a towel or something [67:53] creepy. [laughter] [67:54] >> Yeah. [67:55] >> Creepy robe. Uh [67:58] and uh I think my my last one is just in [68:02] like I'm curious about this kind of [68:04] growing, [68:06] let me see, how do I phrase this? I [68:08] think we're going to see a lot more AI [68:09] hardware, AIdriven hardware, right? And [68:11] a lot of it is going to be built around [68:13] this idea that as as we as users work [68:16] with AI, we realize the value of [68:19] context. I know I do. And so I realize [68:21] the value of having certain [68:22] conversations and and uh meetings and [68:25] whatever recorded so that I can then [68:28] work with that data on a comprehensive [68:30] way after the fact. But what that opens [68:32] is this kind of like growing kind of [68:36] connection into an o a more openly [68:38] surveillent [68:40] world where people want to be recording [68:43] more things. Their devices enable them [68:45] to record more things because they're [68:46] wearing glasses that make it really easy [68:48] or they're wearing a ring that makes it [68:49] really easy. And how how does the how do [68:52] other people in the room feel about that [68:55] weighed with kind of the the the reality [68:58] that having these recordings can be [69:01] really helpful and useful in how we do [69:03] things and so I don't know it's it's a [69:06] weird kind of conundrum state. [69:09] So mine for the next year real quick. Uh [69:11] the first one I'm going to agree with [69:12] your first one which is that I think [69:15] that I'm waiting to see you ask whether [69:17] we're productive and I think at the [69:18] larger level at an industry level. Um is [69:22] there real utility and real revenue [69:24] >> at a scalable constant? Is this really a [69:27] business or not? Right now it's hot and [69:29] people are buying chips and they're [69:30] building centers, but is this really [69:33] going to take off or not? And I think it [69:34] probably will, but uh we'll see whether [69:37] the podcast sticks around as a result, [69:38] too. Number two is kind of coming back [69:41] to the China theme just generally. Will [69:42] Open AAI stumble or will it recover and [69:46] uh and be a real competitor? Right now I [69:48] think they're stumbling. And then number [69:50] three, going back to our earlier [69:52] conversation in the show, um I've been [69:54] thinking about this this last week or so [69:55] and I was thinking that that AI does [69:57] have cooties witness the stories we [69:59] talked about and I wondered whether [70:00] people were going to come to a point [70:01] where they would rebrand AI. They would [70:04] stop calling it AI and they start [70:05] calling other things. So, it so happens [70:07] I this week I listened to a New Books [70:09] Network podcast. It's a podcast of all [70:11] kinds of academic books, really long, [70:13] interesting, but pretty in-depth things. [70:16] Uh, so it was a discussion with Thomas [70:18] Heg, who's a University of Wisconsin, [70:20] Milwaukee professor, um, who's a [70:23] professor, uh, historian of computers [70:25] and technology. He wrote a book about [70:26] the history of computers. He's writing [70:28] one about the history of AI as a brand. [70:32] And in the podcast, interestingly, he [70:34] stole my thunder because he's smarter [70:35] than I am and knows there. And he said [70:37] he predicted that probably this year [70:39] >> coming up a year or so, people will [70:41] start calling it machine learning again [70:44] >> or or other things rather than AI if AI [70:48] takes on enough of a negative [70:49] connotation. [70:50] >> So that'll be interesting to see as we [70:52] go. [70:52] >> Yep. [70:53] >> Absolutely. Although machine learning is [70:55] a little bit more of a a mouthful. [70:57] >> It is [70:58] >> two letters, but yes. No, I think you're [71:01] absolutely right though. I think there [71:02] is there is a little bit of that um that [71:05] cooty factor that that might make that [71:08] need to happen or make people feel like [71:10] it needs to happen. Anyways, [snorts] [71:12] I'm sure we've missed a lot, but that's [71:14] that's a couple of things that we're [71:15] looking towards in the year 2026 outside [71:18] of the fact that we're just going to [71:20] continue doing this show and uh continue [71:22] growing it. And I can't think of a [71:25] person who I'd rather do this with on a [71:27] weekly basis. Jeff, I really appreciate [71:29] doing this show with you. [71:30] >> I learn a ton. I learn from you and uh I [71:33] just appreciate your time. Thank you. [71:35] >> Thank you. [71:36] >> Uh jeffjarvis.com [71:38] is the place where you can go to find [71:40] all of Jeff's works, his books [71:42] primarily. Hot Type coming soon. [71:44] >> Available for pre-order right now. [71:47] >> Right on. Coming soon. Gutenberg [71:49] parenthesis now magazine now and the web [71:51] we weave right now. Jeff Jarvis.com [71:55] and then ainsight.show show for all of [71:57] our episodes, including our previous [72:00] episode where we interviewed Ben [72:01] Girtzel. Again, don't miss that if you [72:03] did already. Uh it's there for you. [72:05] AISight.show. And then finally, [72:07] patreon.comi [72:09] inside. You can go there. You can uh [72:12] follow us and support us on a deeper [72:14] level and we appreciate if you do that. [72:16] And we also have some amazing uh [72:19] executive producers who support us on a [72:21] deeper level each and every week. Dr. [72:24] Jeffrey Marachini, Radio Asheville [72:25] 103.7, Dante St. James, Bono Derrick, [72:28] Jason Knifer, Jason Brady, Anthony DS, [72:31] Mark Starcher, and Karsten Samashki. So, [72:34] thank you everybody, and I'm covering up [72:36] Jeff. I apologize about that. Thank you [72:38] everybody for your support of this show. [72:39] We could not do it without you. I know [72:40] we say that a lot, but it's absolutely [72:42] true. [72:43] >> Um, and then next week, just real quick [72:45] before we end things, I'm going to be in [72:47] Las Vegas for the consumer show. [72:50] >> We didn't talk about that. My Yeah. So, [72:52] we'll we'll talk about that off air. I'm [72:54] I'm guessing probably to next week's [72:56] episode's going to be a little [72:57] different. Like, I might while I'm at [72:59] the show, find a couple of people and do [73:00] some interviews and everything just [73:02] because my schedule is going to be super [73:03] packed. You're going to get an episode. [73:05] Don't know what the format of that is [73:07] going to be yet, but rest assured you'll [73:09] get an episode. And uh it's all there is [73:12] to it. So, thank you everybody. [73:13] >> Stay away from the germs there. Stay [73:15] away from the CES germs, man. [73:16] >> Yeah, I'll have my uh my my hand [73:19] solution. Hopefully, it'll do its job. [73:22] Thank you, Jeff, for a wonderful year [73:23] and looking forward to next year with [73:25] you. And thank you everybody for [73:27] watching and listening. We'll see you [73:28] next time on AI Inside. Happy New Year. [73:32] [music] [73:39] [music]